Full-view point cloud data fixed-length ordered coding method and device and storage medium

Abstract

The invention discloses a full-view point cloud data fixed-length ordered coding method and device, and a storage medium. The method comprises the following steps: obtaining point cloud and performingnormalization processing; constructing a standard coordinate system set containing a plurality of coordinate systems; based on the standard coordinate system set, recoding the point cloud, and calculating and recording obtained intersection point data to obtain fixed-length ordered three-dimensional structure data; and converting the three-dimensional data into fixed-length ordered one-dimensional structure data. The method can encode the point cloud data with any number of points into the point cloud data with a fixed-length structure, and has the advantages of high speed, simple and easy implementation, easy subsequent processing of the encoded data and the like.

Description

technical field [0001] The invention belongs to the fields of computer vision and image processing, and relates to a three-dimensional point cloud encoding method. Background technique [0002] With the wide application of 3D sensing devices, more and more attention has been paid to the acquisition and processing of 3D information of objects, such as 3D reconstruction. Point cloud is the original output of three-dimensional sensing devices such as stereo vision and laser scanners. It is one of the most basic data formats in the field of three-dimensional vision. It is the basis for obtaining three-dimensional information of objects and subsequently reconstructing the surface of objects. Data in this format is usually Record the three-dimensional coordinates of a large number of spatial points on the surface of the observed object in a disordered form. Point cloud data processing is the process of obtaining the category, pose, three-dimensional and other information of objec...

AI Technical Summary

Problems solved by technology

[0004] The first method is to simulate the pixels in the two-dimensional image and convert the point cloud into three-dimensional voxels in space. This method eliminates the disorder of the point cloud, but the computational complexity is still high.

[0005] The second method is to use the relative position relationship of each point in the point cloud data to reorder the points in the original point cloud data, such as k-d tree structure and octree structure, but the data obtained by this method only reflects the distance between points in the point cloud data. The positional relationship of the original point cloud data is not extracted from the global features, and its role in the field of point cloud recognition is limited.

[0007] The deep network pointnet uses the symmetric function max-pooling. After performing a series of the same operations on all points in the point cloud data, the maximum value of the operation results is obtained. It has achieved good results in point cloud classification, but max-pooling will It leads to a large loss of the original information of the point cloud, the subsequent improvement is not effective, and the results are difficult to further improve

The pointnet design of the deep network designed by the Li team of Shandong University uses the x-ray matrix, and the x-ray matrix is ​​multiplied by the original point cloud data to obtain new point cloud data. When the order of the points in the original point cloud data changes, the x-ray The matrix will also change accordingly, so that the obtained point cloud data remains unchanged. This method will not cause loss of point cloud information, but its computational complexity is extremely high, and it is difficult to obtain a perfect x-ray matrix

Images